Earlier this December, Apple made waves when it announced that it will publish its own research papers on artificial intelligence. Within the month, it has announced once again that it published its first AI paper.
The paper titled "Learning from Simulated and Unsupervised Images through Adversarial Training" was published just before Christmas day, December 22 to be exact. It talks about how the use of computer-generated images will improve the ability of an algorithm in recognizing images much better than when using real-life images.
As opposed to real-life images, using computer-generated images in machine learning is more effective because it is already labeled and annotated saving scientists more time and effort.
The paper, however, cautioned that despite the advantage of using synthetic images, there is also a disadvantage in using them because they are not as detailed as real-world images. Thus, the algorithm using computer-generated images will not be able to generalize on real-life images.
In order to solve this problem, the researchers used a new machine learning tool called the Generative Adversial Networks. The tool lets two neural networks work against each other in order to produce photorealistic images. They call this technique Simulated + Unsupervised learning.
With the publication of the paper, Apple is sending a message of becoming more open to the community. For years, the company has been criticized by the tech community for being too secretive. This step is very important since there has been an effort in the whole tech industry to develop a more advanced artificial intelligent software.
The study is headed by Ashish Shrivastava, who has a PhD in computer vision from University of Maryland, College Park. One of the co-authors was Josh Susskind who founded Emotient, an AI start-up which was acquired by Apple this year. Emotient uses a technology which assesses a person's feelings through their expression.
